xref: /linux/arch/mips/kvm/mmu.c (revision 981368e1440b76f68b1ac8f5fb14e739f80ecc4e)
1 /*
2  * This file is subject to the terms and conditions of the GNU General Public
3  * License.  See the file "COPYING" in the main directory of this archive
4  * for more details.
5  *
6  * KVM/MIPS MMU handling in the KVM module.
7  *
8  * Copyright (C) 2012  MIPS Technologies, Inc.  All rights reserved.
9  * Authors: Sanjay Lal <sanjayl@kymasys.com>
10  */
11 
12 #include <linux/highmem.h>
13 #include <linux/kvm_host.h>
14 #include <linux/uaccess.h>
15 #include <asm/mmu_context.h>
16 #include <asm/pgalloc.h>
17 
18 /*
19  * KVM_MMU_CACHE_MIN_PAGES is the number of GPA page table translation levels
20  * for which pages need to be cached.
21  */
22 #if defined(__PAGETABLE_PMD_FOLDED)
23 #define KVM_MMU_CACHE_MIN_PAGES 1
24 #else
25 #define KVM_MMU_CACHE_MIN_PAGES 2
26 #endif
27 
28 void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu)
29 {
30 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
31 }
32 
33 /**
34  * kvm_pgd_init() - Initialise KVM GPA page directory.
35  * @page:	Pointer to page directory (PGD) for KVM GPA.
36  *
37  * Initialise a KVM GPA page directory with pointers to the invalid table, i.e.
38  * representing no mappings. This is similar to pgd_init(), however it
39  * initialises all the page directory pointers, not just the ones corresponding
40  * to the userland address space (since it is for the guest physical address
41  * space rather than a virtual address space).
42  */
43 static void kvm_pgd_init(void *page)
44 {
45 	unsigned long *p, *end;
46 	unsigned long entry;
47 
48 #ifdef __PAGETABLE_PMD_FOLDED
49 	entry = (unsigned long)invalid_pte_table;
50 #else
51 	entry = (unsigned long)invalid_pmd_table;
52 #endif
53 
54 	p = (unsigned long *)page;
55 	end = p + PTRS_PER_PGD;
56 
57 	do {
58 		p[0] = entry;
59 		p[1] = entry;
60 		p[2] = entry;
61 		p[3] = entry;
62 		p[4] = entry;
63 		p += 8;
64 		p[-3] = entry;
65 		p[-2] = entry;
66 		p[-1] = entry;
67 	} while (p != end);
68 }
69 
70 /**
71  * kvm_pgd_alloc() - Allocate and initialise a KVM GPA page directory.
72  *
73  * Allocate a blank KVM GPA page directory (PGD) for representing guest physical
74  * to host physical page mappings.
75  *
76  * Returns:	Pointer to new KVM GPA page directory.
77  *		NULL on allocation failure.
78  */
79 pgd_t *kvm_pgd_alloc(void)
80 {
81 	pgd_t *ret;
82 
83 	ret = (pgd_t *)__get_free_pages(GFP_KERNEL, PGD_TABLE_ORDER);
84 	if (ret)
85 		kvm_pgd_init(ret);
86 
87 	return ret;
88 }
89 
90 /**
91  * kvm_mips_walk_pgd() - Walk page table with optional allocation.
92  * @pgd:	Page directory pointer.
93  * @addr:	Address to index page table using.
94  * @cache:	MMU page cache to allocate new page tables from, or NULL.
95  *
96  * Walk the page tables pointed to by @pgd to find the PTE corresponding to the
97  * address @addr. If page tables don't exist for @addr, they will be created
98  * from the MMU cache if @cache is not NULL.
99  *
100  * Returns:	Pointer to pte_t corresponding to @addr.
101  *		NULL if a page table doesn't exist for @addr and !@cache.
102  *		NULL if a page table allocation failed.
103  */
104 static pte_t *kvm_mips_walk_pgd(pgd_t *pgd, struct kvm_mmu_memory_cache *cache,
105 				unsigned long addr)
106 {
107 	p4d_t *p4d;
108 	pud_t *pud;
109 	pmd_t *pmd;
110 
111 	pgd += pgd_index(addr);
112 	if (pgd_none(*pgd)) {
113 		/* Not used on MIPS yet */
114 		BUG();
115 		return NULL;
116 	}
117 	p4d = p4d_offset(pgd, addr);
118 	pud = pud_offset(p4d, addr);
119 	if (pud_none(*pud)) {
120 		pmd_t *new_pmd;
121 
122 		if (!cache)
123 			return NULL;
124 		new_pmd = kvm_mmu_memory_cache_alloc(cache);
125 		pmd_init(new_pmd);
126 		pud_populate(NULL, pud, new_pmd);
127 	}
128 	pmd = pmd_offset(pud, addr);
129 	if (pmd_none(*pmd)) {
130 		pte_t *new_pte;
131 
132 		if (!cache)
133 			return NULL;
134 		new_pte = kvm_mmu_memory_cache_alloc(cache);
135 		clear_page(new_pte);
136 		pmd_populate_kernel(NULL, pmd, new_pte);
137 	}
138 	return pte_offset_kernel(pmd, addr);
139 }
140 
141 /* Caller must hold kvm->mm_lock */
142 static pte_t *kvm_mips_pte_for_gpa(struct kvm *kvm,
143 				   struct kvm_mmu_memory_cache *cache,
144 				   unsigned long addr)
145 {
146 	return kvm_mips_walk_pgd(kvm->arch.gpa_mm.pgd, cache, addr);
147 }
148 
149 /*
150  * kvm_mips_flush_gpa_{pte,pmd,pud,pgd,pt}.
151  * Flush a range of guest physical address space from the VM's GPA page tables.
152  */
153 
154 static bool kvm_mips_flush_gpa_pte(pte_t *pte, unsigned long start_gpa,
155 				   unsigned long end_gpa)
156 {
157 	int i_min = pte_index(start_gpa);
158 	int i_max = pte_index(end_gpa);
159 	bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PTE - 1);
160 	int i;
161 
162 	for (i = i_min; i <= i_max; ++i) {
163 		if (!pte_present(pte[i]))
164 			continue;
165 
166 		set_pte(pte + i, __pte(0));
167 	}
168 	return safe_to_remove;
169 }
170 
171 static bool kvm_mips_flush_gpa_pmd(pmd_t *pmd, unsigned long start_gpa,
172 				   unsigned long end_gpa)
173 {
174 	pte_t *pte;
175 	unsigned long end = ~0ul;
176 	int i_min = pmd_index(start_gpa);
177 	int i_max = pmd_index(end_gpa);
178 	bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PMD - 1);
179 	int i;
180 
181 	for (i = i_min; i <= i_max; ++i, start_gpa = 0) {
182 		if (!pmd_present(pmd[i]))
183 			continue;
184 
185 		pte = pte_offset_kernel(pmd + i, 0);
186 		if (i == i_max)
187 			end = end_gpa;
188 
189 		if (kvm_mips_flush_gpa_pte(pte, start_gpa, end)) {
190 			pmd_clear(pmd + i);
191 			pte_free_kernel(NULL, pte);
192 		} else {
193 			safe_to_remove = false;
194 		}
195 	}
196 	return safe_to_remove;
197 }
198 
199 static bool kvm_mips_flush_gpa_pud(pud_t *pud, unsigned long start_gpa,
200 				   unsigned long end_gpa)
201 {
202 	pmd_t *pmd;
203 	unsigned long end = ~0ul;
204 	int i_min = pud_index(start_gpa);
205 	int i_max = pud_index(end_gpa);
206 	bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PUD - 1);
207 	int i;
208 
209 	for (i = i_min; i <= i_max; ++i, start_gpa = 0) {
210 		if (!pud_present(pud[i]))
211 			continue;
212 
213 		pmd = pmd_offset(pud + i, 0);
214 		if (i == i_max)
215 			end = end_gpa;
216 
217 		if (kvm_mips_flush_gpa_pmd(pmd, start_gpa, end)) {
218 			pud_clear(pud + i);
219 			pmd_free(NULL, pmd);
220 		} else {
221 			safe_to_remove = false;
222 		}
223 	}
224 	return safe_to_remove;
225 }
226 
227 static bool kvm_mips_flush_gpa_pgd(pgd_t *pgd, unsigned long start_gpa,
228 				   unsigned long end_gpa)
229 {
230 	p4d_t *p4d;
231 	pud_t *pud;
232 	unsigned long end = ~0ul;
233 	int i_min = pgd_index(start_gpa);
234 	int i_max = pgd_index(end_gpa);
235 	bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PGD - 1);
236 	int i;
237 
238 	for (i = i_min; i <= i_max; ++i, start_gpa = 0) {
239 		if (!pgd_present(pgd[i]))
240 			continue;
241 
242 		p4d = p4d_offset(pgd, 0);
243 		pud = pud_offset(p4d + i, 0);
244 		if (i == i_max)
245 			end = end_gpa;
246 
247 		if (kvm_mips_flush_gpa_pud(pud, start_gpa, end)) {
248 			pgd_clear(pgd + i);
249 			pud_free(NULL, pud);
250 		} else {
251 			safe_to_remove = false;
252 		}
253 	}
254 	return safe_to_remove;
255 }
256 
257 /**
258  * kvm_mips_flush_gpa_pt() - Flush a range of guest physical addresses.
259  * @kvm:	KVM pointer.
260  * @start_gfn:	Guest frame number of first page in GPA range to flush.
261  * @end_gfn:	Guest frame number of last page in GPA range to flush.
262  *
263  * Flushes a range of GPA mappings from the GPA page tables.
264  *
265  * The caller must hold the @kvm->mmu_lock spinlock.
266  *
267  * Returns:	Whether its safe to remove the top level page directory because
268  *		all lower levels have been removed.
269  */
270 bool kvm_mips_flush_gpa_pt(struct kvm *kvm, gfn_t start_gfn, gfn_t end_gfn)
271 {
272 	return kvm_mips_flush_gpa_pgd(kvm->arch.gpa_mm.pgd,
273 				      start_gfn << PAGE_SHIFT,
274 				      end_gfn << PAGE_SHIFT);
275 }
276 
277 #define BUILD_PTE_RANGE_OP(name, op)					\
278 static int kvm_mips_##name##_pte(pte_t *pte, unsigned long start,	\
279 				 unsigned long end)			\
280 {									\
281 	int ret = 0;							\
282 	int i_min = pte_index(start);				\
283 	int i_max = pte_index(end);					\
284 	int i;								\
285 	pte_t old, new;							\
286 									\
287 	for (i = i_min; i <= i_max; ++i) {				\
288 		if (!pte_present(pte[i]))				\
289 			continue;					\
290 									\
291 		old = pte[i];						\
292 		new = op(old);						\
293 		if (pte_val(new) == pte_val(old))			\
294 			continue;					\
295 		set_pte(pte + i, new);					\
296 		ret = 1;						\
297 	}								\
298 	return ret;							\
299 }									\
300 									\
301 /* returns true if anything was done */					\
302 static int kvm_mips_##name##_pmd(pmd_t *pmd, unsigned long start,	\
303 				 unsigned long end)			\
304 {									\
305 	int ret = 0;							\
306 	pte_t *pte;							\
307 	unsigned long cur_end = ~0ul;					\
308 	int i_min = pmd_index(start);				\
309 	int i_max = pmd_index(end);					\
310 	int i;								\
311 									\
312 	for (i = i_min; i <= i_max; ++i, start = 0) {			\
313 		if (!pmd_present(pmd[i]))				\
314 			continue;					\
315 									\
316 		pte = pte_offset_kernel(pmd + i, 0);				\
317 		if (i == i_max)						\
318 			cur_end = end;					\
319 									\
320 		ret |= kvm_mips_##name##_pte(pte, start, cur_end);	\
321 	}								\
322 	return ret;							\
323 }									\
324 									\
325 static int kvm_mips_##name##_pud(pud_t *pud, unsigned long start,	\
326 				 unsigned long end)			\
327 {									\
328 	int ret = 0;							\
329 	pmd_t *pmd;							\
330 	unsigned long cur_end = ~0ul;					\
331 	int i_min = pud_index(start);				\
332 	int i_max = pud_index(end);					\
333 	int i;								\
334 									\
335 	for (i = i_min; i <= i_max; ++i, start = 0) {			\
336 		if (!pud_present(pud[i]))				\
337 			continue;					\
338 									\
339 		pmd = pmd_offset(pud + i, 0);				\
340 		if (i == i_max)						\
341 			cur_end = end;					\
342 									\
343 		ret |= kvm_mips_##name##_pmd(pmd, start, cur_end);	\
344 	}								\
345 	return ret;							\
346 }									\
347 									\
348 static int kvm_mips_##name##_pgd(pgd_t *pgd, unsigned long start,	\
349 				 unsigned long end)			\
350 {									\
351 	int ret = 0;							\
352 	p4d_t *p4d;							\
353 	pud_t *pud;							\
354 	unsigned long cur_end = ~0ul;					\
355 	int i_min = pgd_index(start);					\
356 	int i_max = pgd_index(end);					\
357 	int i;								\
358 									\
359 	for (i = i_min; i <= i_max; ++i, start = 0) {			\
360 		if (!pgd_present(pgd[i]))				\
361 			continue;					\
362 									\
363 		p4d = p4d_offset(pgd, 0);				\
364 		pud = pud_offset(p4d + i, 0);				\
365 		if (i == i_max)						\
366 			cur_end = end;					\
367 									\
368 		ret |= kvm_mips_##name##_pud(pud, start, cur_end);	\
369 	}								\
370 	return ret;							\
371 }
372 
373 /*
374  * kvm_mips_mkclean_gpa_pt.
375  * Mark a range of guest physical address space clean (writes fault) in the VM's
376  * GPA page table to allow dirty page tracking.
377  */
378 
379 BUILD_PTE_RANGE_OP(mkclean, pte_mkclean)
380 
381 /**
382  * kvm_mips_mkclean_gpa_pt() - Make a range of guest physical addresses clean.
383  * @kvm:	KVM pointer.
384  * @start_gfn:	Guest frame number of first page in GPA range to flush.
385  * @end_gfn:	Guest frame number of last page in GPA range to flush.
386  *
387  * Make a range of GPA mappings clean so that guest writes will fault and
388  * trigger dirty page logging.
389  *
390  * The caller must hold the @kvm->mmu_lock spinlock.
391  *
392  * Returns:	Whether any GPA mappings were modified, which would require
393  *		derived mappings (GVA page tables & TLB enties) to be
394  *		invalidated.
395  */
396 int kvm_mips_mkclean_gpa_pt(struct kvm *kvm, gfn_t start_gfn, gfn_t end_gfn)
397 {
398 	return kvm_mips_mkclean_pgd(kvm->arch.gpa_mm.pgd,
399 				    start_gfn << PAGE_SHIFT,
400 				    end_gfn << PAGE_SHIFT);
401 }
402 
403 /**
404  * kvm_arch_mmu_enable_log_dirty_pt_masked() - write protect dirty pages
405  * @kvm:	The KVM pointer
406  * @slot:	The memory slot associated with mask
407  * @gfn_offset:	The gfn offset in memory slot
408  * @mask:	The mask of dirty pages at offset 'gfn_offset' in this memory
409  *		slot to be write protected
410  *
411  * Walks bits set in mask write protects the associated pte's. Caller must
412  * acquire @kvm->mmu_lock.
413  */
414 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
415 		struct kvm_memory_slot *slot,
416 		gfn_t gfn_offset, unsigned long mask)
417 {
418 	gfn_t base_gfn = slot->base_gfn + gfn_offset;
419 	gfn_t start = base_gfn +  __ffs(mask);
420 	gfn_t end = base_gfn + __fls(mask);
421 
422 	kvm_mips_mkclean_gpa_pt(kvm, start, end);
423 }
424 
425 /*
426  * kvm_mips_mkold_gpa_pt.
427  * Mark a range of guest physical address space old (all accesses fault) in the
428  * VM's GPA page table to allow detection of commonly used pages.
429  */
430 
431 BUILD_PTE_RANGE_OP(mkold, pte_mkold)
432 
433 static int kvm_mips_mkold_gpa_pt(struct kvm *kvm, gfn_t start_gfn,
434 				 gfn_t end_gfn)
435 {
436 	return kvm_mips_mkold_pgd(kvm->arch.gpa_mm.pgd,
437 				  start_gfn << PAGE_SHIFT,
438 				  end_gfn << PAGE_SHIFT);
439 }
440 
441 bool kvm_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range)
442 {
443 	kvm_mips_flush_gpa_pt(kvm, range->start, range->end);
444 	return true;
445 }
446 
447 bool kvm_set_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
448 {
449 	gpa_t gpa = range->start << PAGE_SHIFT;
450 	pte_t hva_pte = range->arg.pte;
451 	pte_t *gpa_pte = kvm_mips_pte_for_gpa(kvm, NULL, gpa);
452 	pte_t old_pte;
453 
454 	if (!gpa_pte)
455 		return false;
456 
457 	/* Mapping may need adjusting depending on memslot flags */
458 	old_pte = *gpa_pte;
459 	if (range->slot->flags & KVM_MEM_LOG_DIRTY_PAGES && !pte_dirty(old_pte))
460 		hva_pte = pte_mkclean(hva_pte);
461 	else if (range->slot->flags & KVM_MEM_READONLY)
462 		hva_pte = pte_wrprotect(hva_pte);
463 
464 	set_pte(gpa_pte, hva_pte);
465 
466 	/* Replacing an absent or old page doesn't need flushes */
467 	if (!pte_present(old_pte) || !pte_young(old_pte))
468 		return false;
469 
470 	/* Pages swapped, aged, moved, or cleaned require flushes */
471 	return !pte_present(hva_pte) ||
472 	       !pte_young(hva_pte) ||
473 	       pte_pfn(old_pte) != pte_pfn(hva_pte) ||
474 	       (pte_dirty(old_pte) && !pte_dirty(hva_pte));
475 }
476 
477 bool kvm_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
478 {
479 	return kvm_mips_mkold_gpa_pt(kvm, range->start, range->end);
480 }
481 
482 bool kvm_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
483 {
484 	gpa_t gpa = range->start << PAGE_SHIFT;
485 	pte_t *gpa_pte = kvm_mips_pte_for_gpa(kvm, NULL, gpa);
486 
487 	if (!gpa_pte)
488 		return false;
489 	return pte_young(*gpa_pte);
490 }
491 
492 /**
493  * _kvm_mips_map_page_fast() - Fast path GPA fault handler.
494  * @vcpu:		VCPU pointer.
495  * @gpa:		Guest physical address of fault.
496  * @write_fault:	Whether the fault was due to a write.
497  * @out_entry:		New PTE for @gpa (written on success unless NULL).
498  * @out_buddy:		New PTE for @gpa's buddy (written on success unless
499  *			NULL).
500  *
501  * Perform fast path GPA fault handling, doing all that can be done without
502  * calling into KVM. This handles marking old pages young (for idle page
503  * tracking), and dirtying of clean pages (for dirty page logging).
504  *
505  * Returns:	0 on success, in which case we can update derived mappings and
506  *		resume guest execution.
507  *		-EFAULT on failure due to absent GPA mapping or write to
508  *		read-only page, in which case KVM must be consulted.
509  */
510 static int _kvm_mips_map_page_fast(struct kvm_vcpu *vcpu, unsigned long gpa,
511 				   bool write_fault,
512 				   pte_t *out_entry, pte_t *out_buddy)
513 {
514 	struct kvm *kvm = vcpu->kvm;
515 	gfn_t gfn = gpa >> PAGE_SHIFT;
516 	pte_t *ptep;
517 	kvm_pfn_t pfn = 0;	/* silence bogus GCC warning */
518 	bool pfn_valid = false;
519 	int ret = 0;
520 
521 	spin_lock(&kvm->mmu_lock);
522 
523 	/* Fast path - just check GPA page table for an existing entry */
524 	ptep = kvm_mips_pte_for_gpa(kvm, NULL, gpa);
525 	if (!ptep || !pte_present(*ptep)) {
526 		ret = -EFAULT;
527 		goto out;
528 	}
529 
530 	/* Track access to pages marked old */
531 	if (!pte_young(*ptep)) {
532 		set_pte(ptep, pte_mkyoung(*ptep));
533 		pfn = pte_pfn(*ptep);
534 		pfn_valid = true;
535 		/* call kvm_set_pfn_accessed() after unlock */
536 	}
537 	if (write_fault && !pte_dirty(*ptep)) {
538 		if (!pte_write(*ptep)) {
539 			ret = -EFAULT;
540 			goto out;
541 		}
542 
543 		/* Track dirtying of writeable pages */
544 		set_pte(ptep, pte_mkdirty(*ptep));
545 		pfn = pte_pfn(*ptep);
546 		mark_page_dirty(kvm, gfn);
547 		kvm_set_pfn_dirty(pfn);
548 	}
549 
550 	if (out_entry)
551 		*out_entry = *ptep;
552 	if (out_buddy)
553 		*out_buddy = *ptep_buddy(ptep);
554 
555 out:
556 	spin_unlock(&kvm->mmu_lock);
557 	if (pfn_valid)
558 		kvm_set_pfn_accessed(pfn);
559 	return ret;
560 }
561 
562 /**
563  * kvm_mips_map_page() - Map a guest physical page.
564  * @vcpu:		VCPU pointer.
565  * @gpa:		Guest physical address of fault.
566  * @write_fault:	Whether the fault was due to a write.
567  * @out_entry:		New PTE for @gpa (written on success unless NULL).
568  * @out_buddy:		New PTE for @gpa's buddy (written on success unless
569  *			NULL).
570  *
571  * Handle GPA faults by creating a new GPA mapping (or updating an existing
572  * one).
573  *
574  * This takes care of marking pages young or dirty (idle/dirty page tracking),
575  * asking KVM for the corresponding PFN, and creating a mapping in the GPA page
576  * tables. Derived mappings (GVA page tables and TLBs) must be handled by the
577  * caller.
578  *
579  * Returns:	0 on success, in which case the caller may use the @out_entry
580  *		and @out_buddy PTEs to update derived mappings and resume guest
581  *		execution.
582  *		-EFAULT if there is no memory region at @gpa or a write was
583  *		attempted to a read-only memory region. This is usually handled
584  *		as an MMIO access.
585  */
586 static int kvm_mips_map_page(struct kvm_vcpu *vcpu, unsigned long gpa,
587 			     bool write_fault,
588 			     pte_t *out_entry, pte_t *out_buddy)
589 {
590 	struct kvm *kvm = vcpu->kvm;
591 	struct kvm_mmu_memory_cache *memcache = &vcpu->arch.mmu_page_cache;
592 	gfn_t gfn = gpa >> PAGE_SHIFT;
593 	int srcu_idx, err;
594 	kvm_pfn_t pfn;
595 	pte_t *ptep, entry, old_pte;
596 	bool writeable;
597 	unsigned long prot_bits;
598 	unsigned long mmu_seq;
599 
600 	/* Try the fast path to handle old / clean pages */
601 	srcu_idx = srcu_read_lock(&kvm->srcu);
602 	err = _kvm_mips_map_page_fast(vcpu, gpa, write_fault, out_entry,
603 				      out_buddy);
604 	if (!err)
605 		goto out;
606 
607 	/* We need a minimum of cached pages ready for page table creation */
608 	err = kvm_mmu_topup_memory_cache(memcache, KVM_MMU_CACHE_MIN_PAGES);
609 	if (err)
610 		goto out;
611 
612 retry:
613 	/*
614 	 * Used to check for invalidations in progress, of the pfn that is
615 	 * returned by pfn_to_pfn_prot below.
616 	 */
617 	mmu_seq = kvm->mmu_invalidate_seq;
618 	/*
619 	 * Ensure the read of mmu_invalidate_seq isn't reordered with PTE reads
620 	 * in gfn_to_pfn_prot() (which calls get_user_pages()), so that we don't
621 	 * risk the page we get a reference to getting unmapped before we have a
622 	 * chance to grab the mmu_lock without mmu_invalidate_retry() noticing.
623 	 *
624 	 * This smp_rmb() pairs with the effective smp_wmb() of the combination
625 	 * of the pte_unmap_unlock() after the PTE is zapped, and the
626 	 * spin_lock() in kvm_mmu_notifier_invalidate_<page|range_end>() before
627 	 * mmu_invalidate_seq is incremented.
628 	 */
629 	smp_rmb();
630 
631 	/* Slow path - ask KVM core whether we can access this GPA */
632 	pfn = gfn_to_pfn_prot(kvm, gfn, write_fault, &writeable);
633 	if (is_error_noslot_pfn(pfn)) {
634 		err = -EFAULT;
635 		goto out;
636 	}
637 
638 	spin_lock(&kvm->mmu_lock);
639 	/* Check if an invalidation has taken place since we got pfn */
640 	if (mmu_invalidate_retry(kvm, mmu_seq)) {
641 		/*
642 		 * This can happen when mappings are changed asynchronously, but
643 		 * also synchronously if a COW is triggered by
644 		 * gfn_to_pfn_prot().
645 		 */
646 		spin_unlock(&kvm->mmu_lock);
647 		kvm_release_pfn_clean(pfn);
648 		goto retry;
649 	}
650 
651 	/* Ensure page tables are allocated */
652 	ptep = kvm_mips_pte_for_gpa(kvm, memcache, gpa);
653 
654 	/* Set up the PTE */
655 	prot_bits = _PAGE_PRESENT | __READABLE | _page_cachable_default;
656 	if (writeable) {
657 		prot_bits |= _PAGE_WRITE;
658 		if (write_fault) {
659 			prot_bits |= __WRITEABLE;
660 			mark_page_dirty(kvm, gfn);
661 			kvm_set_pfn_dirty(pfn);
662 		}
663 	}
664 	entry = pfn_pte(pfn, __pgprot(prot_bits));
665 
666 	/* Write the PTE */
667 	old_pte = *ptep;
668 	set_pte(ptep, entry);
669 
670 	err = 0;
671 	if (out_entry)
672 		*out_entry = *ptep;
673 	if (out_buddy)
674 		*out_buddy = *ptep_buddy(ptep);
675 
676 	spin_unlock(&kvm->mmu_lock);
677 	kvm_release_pfn_clean(pfn);
678 	kvm_set_pfn_accessed(pfn);
679 out:
680 	srcu_read_unlock(&kvm->srcu, srcu_idx);
681 	return err;
682 }
683 
684 int kvm_mips_handle_vz_root_tlb_fault(unsigned long badvaddr,
685 				      struct kvm_vcpu *vcpu,
686 				      bool write_fault)
687 {
688 	int ret;
689 
690 	ret = kvm_mips_map_page(vcpu, badvaddr, write_fault, NULL, NULL);
691 	if (ret)
692 		return ret;
693 
694 	/* Invalidate this entry in the TLB */
695 	return kvm_vz_host_tlb_inv(vcpu, badvaddr);
696 }
697 
698 /**
699  * kvm_mips_migrate_count() - Migrate timer.
700  * @vcpu:	Virtual CPU.
701  *
702  * Migrate CP0_Count hrtimer to the current CPU by cancelling and restarting it
703  * if it was running prior to being cancelled.
704  *
705  * Must be called when the VCPU is migrated to a different CPU to ensure that
706  * timer expiry during guest execution interrupts the guest and causes the
707  * interrupt to be delivered in a timely manner.
708  */
709 static void kvm_mips_migrate_count(struct kvm_vcpu *vcpu)
710 {
711 	if (hrtimer_cancel(&vcpu->arch.comparecount_timer))
712 		hrtimer_restart(&vcpu->arch.comparecount_timer);
713 }
714 
715 /* Restore ASID once we are scheduled back after preemption */
716 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
717 {
718 	unsigned long flags;
719 
720 	kvm_debug("%s: vcpu %p, cpu: %d\n", __func__, vcpu, cpu);
721 
722 	local_irq_save(flags);
723 
724 	vcpu->cpu = cpu;
725 	if (vcpu->arch.last_sched_cpu != cpu) {
726 		kvm_debug("[%d->%d]KVM VCPU[%d] switch\n",
727 			  vcpu->arch.last_sched_cpu, cpu, vcpu->vcpu_id);
728 		/*
729 		 * Migrate the timer interrupt to the current CPU so that it
730 		 * always interrupts the guest and synchronously triggers a
731 		 * guest timer interrupt.
732 		 */
733 		kvm_mips_migrate_count(vcpu);
734 	}
735 
736 	/* restore guest state to registers */
737 	kvm_mips_callbacks->vcpu_load(vcpu, cpu);
738 
739 	local_irq_restore(flags);
740 }
741 
742 /* ASID can change if another task is scheduled during preemption */
743 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
744 {
745 	unsigned long flags;
746 	int cpu;
747 
748 	local_irq_save(flags);
749 
750 	cpu = smp_processor_id();
751 	vcpu->arch.last_sched_cpu = cpu;
752 	vcpu->cpu = -1;
753 
754 	/* save guest state in registers */
755 	kvm_mips_callbacks->vcpu_put(vcpu, cpu);
756 
757 	local_irq_restore(flags);
758 }
759